Ba­sics of AI

Ar­ti­fi­cial In­tel­li­gence, Ma­chine Learn­ing and Deep Learn­ing are all buzz words that seem to be ev­ery­where th­ese days. They’re in your smart­phone, your smart­watch, your smart speak­ers and smart fridge. But what’s the dif­fer­ence be­tween them, and are they r

HWM (Singapore) - - CONTENTS - By James Lu


Ar­ti­fi­cial In­tel­li­gence is a gen­eral term for any form of com­puter think­ing. It can re­fer to any­thing from the AI op­po­nent in a game of Star­craft, to a voice-recog­ni­tion sys­tem like Siri or Google Now in­ter­pret­ing and re­spond­ing to speech.

Fur­ther­more, the tech­nol­ogy can broadly be cat­e­go­rized into two groups: Nar­row AI and Ar­ti­fi­cial Gen­eral In­tel­li­gence (AGI).

Google’s Deep­Mind Al­phaGo AI is an ex­am­ple of nar­row AI: one that is skilled at a spe­cific task. In con­trast, an AGI is a gen­eral pur­pose AI that, in the­ory can ap­ply its think­ing to any task. A true AGI is much harder to cre­ate and while there have been many at­tempts to make one, so far, no AI can be clas­si­fied as an AGI.


Ma­chine learn­ing is a sub­set of AI. It refers specif­i­cally to soft­ware de­signed to de­tect pat­terns and ob­serve out­comes, then use that anal­y­sis to ad­just its own be­hav­ior. Ma­chine learn­ing doesn’t ac­tu­ally re­quire in­tel­li­gent think­ing in the way we per­ceive it, it sim­ply re­quires re­ally good pat­tern match­ing and the abil­ity to ap­ply those pat­terns to its be­hav­ior.

IBM’s Deep Blue and Deep­Mind’s Al­pha Go are both game-play­ing Nar­row AIs, but only Al­pha Go uses Ma­chine Learn­ing. Deep Blue uses rule-based pro­gram­ming, so any changes in its be­hav­ior re­lies on changes in its core pro­gram­ming. On the other hand, Al­pha Go was able to beat Go world cham­pion Ke Jie by an­a­lyz­ing ex­pert-level Go matches and ap­ply­ing those strate­gies.


Deep Learn­ing is a fur­ther sub­set of Ma­chine Learn­ing that uses al­go­rithms in­spired by the struc­ture of the hu­man brain called ar­ti­fi­cial neu­ral net­works to solve prob­lems. In Deep Learn­ing, the AI have mul­ti­ple lay­ers that han­dle dif­fer­ent spe­cific tasks. And even some­thing as sim­ple as iden­ti­fy­ing a stop sign will have dif­fer­ent lay­ers to an­a­lyze shape, color, pat­terns, and text. As such, Deep Learn­ing re­quires mas­sive datasets to work. For ex­am­ple, Google’s self-driv­ing cars are trained to rec­og­nize ob­sta­cles and re­act to them ap­pro­pri­ately. But due to the in­fi­nite num­ber of vari­ables (other cars, pedes­tri­ans, weather and road con­di­tions etc.) in­volved, Google re­quires a mas­sive amount of data to an­a­lyze.

Newspapers in English

Newspapers from Singapore

© PressReader. All rights reserved.